Motphys Products

Rigidbody physics simulation covers features such as collision detection, dynamics, joints, and scene queries for rigid bodies. To better meet the needs of industrial simulation, we further support generalized coordinates and multi-body physics modeling to achieve more precise drive models and superior dynamic stability. Additionally, in the field of providing physical simulation environments for machine learning, we have also launched a GPU version of the rigidbody physics engine to accelerate training efficiency, and support parallel simulation of multiple worlds.

Leveraging the parallel computing advantages of GPUs, we can rapidly simulate the dynamic behavior and interaction of large-scale particles, thereby reproducing various complex physical phenomena in the real world. The unified solving framework can handle multiple physical materials including cloth, hair, soft bodies, and fluids, and supports interaction effects between these different materials. Shader translation and RHI cross-platform design allow for compilation and deployment on various GPU platforms, including GPUs of various mobile devices.

Distributed computing for real-time motion physics simulation has successfully solved the computational bottleneck problem encountered in large-scale physics simulation on a single machine. With this technology, we can build large-scale cloud-based physical simulation environments to provide powerful support for large-scale parallel AI model training; on the other hand, as a vast virtual world, the metaverse will have a large number of participants, forming a scenario of large-scale user real-time interaction, requiring ultra-large-scale physical simulation, which is now possible.

We apply physical simulation technology to the field of artificial intelligence and have deeply integrated with PyTorch and Gymnasium. Using reinforcement learning and imitation learning algorithms, we have the ability to train embodied intelligence and characters in virtual physical simulation environments. The advantage of the physics engine is that it can simulate various complex physical phenomena, thereby generating a large amount of training data; moreover, efficient and stable physical simulation environments can significantly shorten training time, thus saving a lot of computational resources.